Lachance Alexandre, Godbout Mathieu, Antaki Fares, Hébert Mélanie, Bourgault Serge, Caissie Mathieu, Tourville Éric, Durand Audrey, Dirani Ali
Faculté de Médecine, Université Laval, Québec, QC, Canada.
Département d'Ophtalmologie et d'oto-Rhino-Laryngologie - Chirurgie Cervico-Faciale, Centre Universitaire d'Ophtalmologie, Hôpital du Saint-Sacrement, CHU de Québec - Université Laval, Québec, QC, Canada.
Transl Vis Sci Technol. 2022 Apr 1;11(4):6. doi: 10.1167/tvst.11.4.6.
The purpose of this study was to assess the feasibility of deep learning (DL) methods to enhance the prediction of visual acuity (VA) improvement after macular hole (MH) surgery from a combined model using DL on high-definition optical coherence tomography (HD-OCT) B-scans and clinical features.
We trained a DL convolutional neural network (CNN) using pre-operative HD-OCT B-scans of the macula and combined with a logistic regression model of pre-operative clinical features to predict VA increase ≥15 Early Treatment Diabetic Retinopathy Study (ETDRS) letters at 6 months post-vitrectomy in closed MHs. A total of 121 MHs with 242 HD-OCT B-scans and 484 clinical data points were used to train, validate, and test the model. Prediction of VA increase was evaluated using the area under the receiver operating characteristic curve (AUROC) and F1 scores. We also extracted the weight of each input feature in the hybrid model.
All performances are reported on the held-out test set, matching results obtained with cross-validation. Using a regression on clinical features, the AUROC was 80.6, with an F1 score of 79.7. For the CNN, relying solely on the HD-OCT B-scans, the AUROC was 72.8 ± 14.6, with a F1 score of 61.5 ± 23.7. For our hybrid regression model using clinical features and CNN prediction, the AUROC was 81.9 ± 5.2, with an F1 score of 80.4 ± 7.7. In the hybrid model, the baseline VA was the most important feature (weight = 59.1 ± 6.9%), while the weight of HD-OCT prediction was 9.6 ± 4.2%.
Both the clinical data and HD-OCT models can predict postoperative VA improvement in patients undergoing vitrectomy for a MH with good discriminative performances. Combining them into a hybrid model did not significantly improve performance.
OCT-based DL models can predict postoperative VA improvement following vitrectomy for MH but fusing those models with clinical data might not provide improved predictive performance.
本研究旨在评估深度学习（DL）方法的可行性，该方法通过使用深度学习对高清光学相干断层扫描（HD-OCT）B 扫描和临床特征构建的组合模型，来增强黄斑裂孔（MH）手术后视力（VA）改善情况的预测。
我们使用黄斑术前 HD-OCT B 扫描训练了一个 DL 卷积神经网络（CNN），并结合术前临床特征的逻辑回归模型，以预测闭合性 MH 玻璃体切除术后 6 个月时 VA 提高≥15 个早期糖尿病视网膜病变研究（ETDRS）字母。总共 121 个 MH，242 次 HD-OCT B 扫描和 484 个临床数据点用于训练、验证和测试该模型。使用受试者工作特征曲线下面积（AUROC）和 F1 分数评估 VA 提高的预测情况。我们还提取了混合模型中每个输入特征的权重。
所有性能均在保留测试集上报告，与交叉验证获得的结果匹配。使用临床特征进行回归分析时，AUROC 为 80.6，F1 分数为 79.7。对于仅依靠 HD-OCT B 扫描的 CNN，AUROC 为 72.8±14.6，F1 分数为 61.5±23.7。对于我们使用临床特征和 CNN 预测的混合回归模型，AUROC 为 81.9±5.2，F1 分数为 80.4±7.7。在混合模型中，基线 VA 是最重要的特征（权重 = 59.1±6.9%），而 HD-OCT 预测的权重为 9.6±4.2%。
临床数据模型和 HD-OCT 模型均能对接受 MH 玻璃体切除术患者的术后 VA 改善情况进行预测，且具有良好的判别性能。将它们组合成一个混合模型并没有显著提高性能。
基于 OCT 的 DL 模型可以预测 MH 玻璃体切除术后的 VA 改善情况，但将这些模型与临床数据融合可能无法提高预测性能。
